This technology is a wireless communication network packet scheduling method and apparatus that builds a machine learning-based Proportional Fair Scheduling (PFS) channel allocation model using channel information, user average throughput up to the previous time slot, and selected users from previous time slots as training data, and determines user-channel allocation by inputting test data.
Proportional Fair Scheduling is widely used due to its excellent balance between system throughput and fairness; however, most existing packet scheduling solutions rely on prescribed algorithms or iterative calculations, leading to high computational overhead. While there have been attempts to introduce machine learning into network resource allocation, there were limitations in that the feasibility and effectiveness of applying it to specific scheduling algorithms had not been sufficiently verified.
This technology features a performance evaluation loop that learns PFS allocation results using a polynomial support vector machine and retrains the model by adjusting the polynomial degree if the allocation latency exceeds a threshold. This allows for efficient user-to-channel allocation with low computational overhead, eliminating the need for repetitive calculations every slot. It can be applied to LTE/5G base station schedulers, multi-carrier cellular systems, and edge computing-based wireless resource management, reducing both scheduling latency and base station computational load while maintaining fairness.
This invention was developed with support from the Ministry of Education's research on learning-based fog network caching optimization for ultra-reliable low-latency services.
N/A
